The basic concept of personalized medicine is to tailor the treatment for a patient based on his or her genetic makeup, clinical conditions and other personal characteristics to improve efficacy and safety. The coming of the big data era enables us to characterize individual in fine pictures and make the personalized clinical decision truly personalized. Such an approach has great potential for improving disease prevention, diagnosis and treatment. For example, in a typical randomized clinical trial aiming for proving the efficacy of a treatment, the final conclusion is drawn based on the average treatment effect in the entire study population. It is possible that while the average treatment effect is near null, the treatment may still be beneficial to a subgroup of patients whose identification prior to the treatment is thus very important. The overall objective of statistical analysis in this area is to provide a data-based empirical estimator for the personalized treatment effect, which can be used to identify subgroup of patients who may benefit the most from a treatment. In this study, we first propose to develop robust statistical methods for estimating the group-specific treatment effect. The proposed approach incorporating many existing methods as special cases depends on minimum model assumptions and provides a general framework for generalization and improvement. We will also discuss how to use the estimated personalized treatment effect to stratify patient population into clinically meaningful strata for better assisting the decision making of clinicians. Secondly, we will study a regularize principal components analysis method for dimension reduction in structured high-dimensional data. The output from the analysis can be used to summarize the characteristics of individual patient as well as for predicting future clinical outcomes of interest. Multiple methods can be used to estimate the treatment effect and form the corresponding treatment selection strategy. Therefore it is important to evaluate and compare the performance of such strategies. Thus our last aim is to develop a systematic robust procedure for evaluating the performance of the personalized treatment effect estimation and associated treatment selection strategy.